The Problem Machine Learning Can’t Learn Is a Problem
Machine learning and deep learning were built by engineers and coders to give our machines the ability to learn on their own. Given a dataset and a goal, the machine is left to work out the best possible answer for itself. This capability has driven a near-quantum leap in technology over the last decade. And as our devices grow smarter, it stands to reason that we, their beneficiaries, grow more dependent on them.
Some examples are already evident. Thirty years ago, when rotary phones were plugged into the wall at home, we memorized the numbers we called most. Today our cell phones store every contact under a name, and many of us can no longer recall those numbers. On its own, this is hardly a catastrophe. But consider what happens when we automate the entire process of producing the devices we depend on. Within a short time, the know-how of making those machines could fade from human memory, leaving us wholly reliant on machines to build the tools we need. What happens when, as a species, we lose the knowledge to repair, troubleshoot, and program AI itself?
This is a problem many people don’t realize exists, and one that cannot be taught to AI to solve for us. History has shown repeatedly that once a productive class takes over the work of a society, its beneficiaries lose nearly all knowledge of doing for themselves. AI is poised to become that class. This is the problem machine learning can’t learn. We can train a machine to take over tasks humans once did themselves, but we can never train it to grasp what that help does to us: every task it automates makes us a little less inclined to seek out the knowledge to do it ourselves. A machine optimizes toward the goals we give it, so it has no stake in whether humans keep their own skills. Its success is measured in tasks completed, not in capability preserved. To the machine, a dependent human and a capable human look exactly the same.
History has shown us how quietly this happens. Aviation regulators and safety investigators have warned for years that heavy reliance on autopilot is eroding pilots’ manual flying skills, leaving some less prepared for the rare moment when the automation fails and a human must take over. Navigation tells a similar story. Before GPS, drivers built mental maps of their cities and learned landmarks and routes; today many of us can’t find our way across town without a voice telling us where to turn, and research has linked heavy GPS use to weaker spatial memory. Each of these tools was a genuine improvement, and each left us a little less able to manage without it.
Defenders of automation will say we have always offloaded tasks to our tools, and that is true. But AI is different in a way that matters. Every earlier tool still needed a human hand on it. The pilot still monitored the autopilot and the driver still chose the destination. We were not setting it and forgetting it. With AI, that is increasingly exactly what we are doing, and we will likely reach a point where we trust it enough to make it the voice of reason, and even the arbiter of truth. A civilization that can no longer build, repair, or check its machines, and that has handed those machines the final word on what is true, has no way left to tell when they are wrong.
That forces us to ask what we, as a species, will hold sacred. Will there be sectors, industries, and parts of life that we forbid AI to enter? Will we let technology run free in the name of progress and convenience? Or will we accept a duty of self-preservation: to hone, maintain, and capitalize on the parts of our knowledge and consciousness in which machines should have no part?
My personal opinion, pertaining to the storytelling business I don’t believe machines should ever have a part in or be granted the ability to create author or dictate stories, they should be completely banned from the recreation of history as it applies to man and should have no part in any aspect of artistic expression. This is not about protecting jobs. It is about ensuring our history is never rewritten in tellings unfavorable to us, and that we are never written out of our own story. Machines must not decide what counts as great art, because their influence alone could set a standard that humans eventually cannot meet. Finally, AI, robots, and their kind should never hold positions of authority that require them to decide matters of life and death.